fix: api keys form bugfix and org key sweeper job

Signed-off-by: allanice001 <allanice001@gmail.com>

internal/bg/bg.go

@@ -129,6 +129,12 @@ func NewJobs(gdb *gorm.DB, dbUrl string) (*Jobs, error) {
 		archer.WithTimeout(60*time.Minute),
 	)
 
+	c.Register(
+		"org_key_sweeper",
+		OrgKeySweeperWorker(gdb, jobs),
+		archer.WithInstances(1),
+		archer.WithTimeout(5*time.Minute),
+	)
 	return jobs, nil
 }
 

internal/bg/org_key_sweeper.go

@@ -0,0 +1,95 @@
+package bg
+
+import (
+	"context"
+	"time"
+
+	"github.com/dyaksa/archer"
+	"github.com/dyaksa/archer/job"
+	"github.com/glueops/autoglue/internal/models"
+	"github.com/google/uuid"
+	"github.com/rs/zerolog/log"
+	"gorm.io/gorm"
+)
+
+type OrgKeySweeperArgs struct {
+	IntervalS     int `json:"interval_seconds,omitempty"`
+	RetentionDays int `json:"retention_days,omitempty"`
+}
+
+type OrgKeySweeperResult struct {
+	Status           string `json:"status"`
+	MarkedRevoked    int    `json:"marked_revoked"`
+	DeletedEphemeral int    `json:"deleted_ephemeral"`
+	ElapsedMs        int    `json:"elapsed_ms"`
+}
+
+func OrgKeySweeperWorker(db *gorm.DB, jobs *Jobs) archer.WorkerFn {
+	return func(ctx context.Context, j job.Job) (any, error) {
+		args := OrgKeySweeperArgs{
+			IntervalS:     3600,
+			RetentionDays: 10,
+		}
+		start := time.Now()
+
+		_ = j.ParseArguments(&args)
+		if args.IntervalS <= 0 {
+			args.IntervalS = 3600
+		}
+		if args.RetentionDays <= 0 {
+			args.RetentionDays = 10
+		}
+
+		now := time.Now()
+
+		// 1) Mark expired keys as revoked
+		res1 := db.Model(&models.APIKey{}).
+			Where("expires_at IS NOT NULL AND expires_at <= ? AND revoked = false", now).
+			Updates(map[string]any{
+				"revoked":    true,
+				"updated_at": now,
+			})
+
+		if res1.Error != nil {
+			log.Error().Err(res1.Error).Msg("[org_key_sweeper] mark expired revoked failed")
+			return nil, res1.Error
+		}
+		markedRevoked := int(res1.RowsAffected)
+
+		// 2) Hard-delete ephemeral keys that are revoked and older than retention
+		cutoff := now.Add(-time.Duration(args.RetentionDays) * 24 * time.Hour)
+		res2 := db.
+			Where("is_ephemeral = ? AND revoked = ? AND updated_at <= ?", true, true, cutoff).
+			Delete(&models.APIKey{})
+
+		if res2.Error != nil {
+			log.Error().Err(res2.Error).Msg("[org_key_sweeper] delete revoked ephemeral keys failed")
+			return nil, res2.Error
+		}
+		deletedEphemeral := int(res2.RowsAffected)
+
+		out := OrgKeySweeperResult{
+			Status:           "ok",
+			MarkedRevoked:    markedRevoked,
+			DeletedEphemeral: deletedEphemeral,
+			ElapsedMs:        int(time.Since(start).Milliseconds()),
+		}
+
+		log.Info().
+			Int("marked_revoked", markedRevoked).
+			Int("deleted_ephemeral", deletedEphemeral).
+			Msg("[org_key_sweeper] cleanup tick ok")
+
+		// Re-enqueue the sweeper
+		next := time.Now().Add(time.Duration(args.IntervalS) * time.Second)
+		_, _ = jobs.Enqueue(
+			ctx,
+			uuid.NewString(),
+			"org_key_sweeper",
+			args,
+			archer.WithScheduleTime(next),
+			archer.WithMaxRetries(1),
+		)
+		return out, nil
+	}
+}

internal/bg/prepare_cluster.go

@@ -158,10 +158,11 @@ func ClusterPrepareWorker(db *gorm.DB, jobs *Jobs) archer.WorkerFn {
 				dtoCluster.Kubeconfig = &kubeconfig
 			}
 
-			orgKey, orgSecret, err := createOrgScopedKeyForPayload(
+			orgKey, orgSecret, err := findOrCreateClusterAutomationKey(
 				db,
 				c.OrganizationID,
-				fmt.Sprintf("cluster-%s-%s", c.Name, c.ID.String()),
+				c.ID,
+				24*time.Hour,
 			)
 
 			if err != nil {
@@ -584,12 +585,28 @@ func randomB64URL(n int) (string, error) {
 	return base64.RawURLEncoding.EncodeToString(b), nil
 }
 
-func createOrgScopedKeyForPayload(db *gorm.DB, orgID uuid.UUID, name string) (orgKey string, orgSecret string, err error) {
+func findOrCreateClusterAutomationKey(
+	db *gorm.DB,
+	orgID uuid.UUID,
+	clusterID uuid.UUID,
+	ttl time.Duration,
+) (orgKey string, orgSecret string, err error) {
+	now := time.Now()
+	name := fmt.Sprintf("cluster-%s-bastion", clusterID.String())
+
+	// 1) Delete any existing ephemeral cluster-bastion key for this org+cluster
+	if err := db.Where(
+		"org_id = ? AND scope = ? AND purpose = ? AND cluster_id = ? AND is_ephemeral = ?",
+		orgID, "org", "cluster_bastion", clusterID, true,
+	).Delete(&models.APIKey{}).Error; err != nil {
+		return "", "", fmt.Errorf("delete existing cluster key: %w", err)
+	}
+
+	// 2) Mint a fresh keypair
 	keySuffix, err := randomB64URL(16)
 	if err != nil {
 		return "", "", fmt.Errorf("entropy_error: %w", err)
 	}
-
 	sec, err := randomB64URL(32)
 	if err != nil {
 		return "", "", fmt.Errorf("entropy_error: %w", err)
@@ -604,13 +621,25 @@ func createOrgScopedKeyForPayload(db *gorm.DB, orgID uuid.UUID, name string) (or
 		return "", "", fmt.Errorf("hash_error: %w", err)
 	}
 
+	exp := now.Add(ttl)
+
+	prefix := orgKey
+	if len(prefix) > 12 {
+		prefix = prefix[:12]
+	}
+
 	rec := models.APIKey{
-		OrgID:      &orgID,
-		Scope:      "org",
-		Name:       name,
-		KeyHash:    keyHash,
-		SecretHash: &secretHash,
-		ExpiresAt:  nil,
+		OrgID:       &orgID,
+		Scope:       "org",
+		Purpose:     "cluster_bastion",
+		ClusterID:   &clusterID,
+		IsEphemeral: true,
+		Name:        name,
+		KeyHash:     keyHash,
+		SecretHash:  &secretHash,
+		ExpiresAt:   &exp,
+		Revoked:     false,
+		Prefix:      &prefix,
 	}
 
 	if err := db.Create(&rec).Error; err != nil {